Skip to content

Matmul - Add More Tests for 2D interleaved in0 / batched height sharded in1 #37789

@edwinleeTT

Description

@edwinleeTT

#37681 adds support for the 2D matmul to run DRAM matmuls with interleaved activations and batched height sharded weights. Unit tests were added to exercise the two MLA prefill matmuls that required this support. However, we should add more test cases to nightly to cover a wider range of inputs.

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions